Numaflow Perfman

Running Perfman

Perfman assumes your system already has a local kubernetes cluster with at least a single node configured. Perfman can be run in two different ways.

With an executable:

1. Clone the repo and make it the active working directory: git clone git@github.com:numaproj-labs/numaflow-perfman.git && cd numaflow-perfman
2. Run make build. This will produce an executable: dist/perfman
3. Use perfman with desired commands: ./dist/perfman help

Note: if you use your current terminal session to port-forward, since this is a blocking operation, open a new terminal tab to run new commands.

With Docker:

1. Pull the stable image from registry: docker pull quay.io/numaio/numaproj-labs/perfman:stable
2. Run the container with mkdir -p output && docker run -it --network host -v ~/.kube/config:/perfmanuser/.kube/config:ro -v ./output:/home/perfman/output quay.io/numaio/numaproj-labs/perfman:stable (if you have the repo cloned, this can be accomplished with make run). This assumes that the local kubernetes configuration file for your cluster is located at $HOME/.kube/config
3. The above command will spawn a shell inside the container, and perfman can be used with the desired commands: perfman help

Note: As above, port-forwarding is a blocking operation, and with the Docker option, open a new terminal tab and:

• Identify the name/ID of the currently running perfman container with docker ps
• Run docker exec -it <name/ID> /bin/ash. This will provide a new terminal session inside the same container, where you can run additional commands

Collecting Metrics

The following steps can be followed in order to collect metrics for the base perfman pipeline. Note that the metrics are outputted as CSV files under the output/ directory. The following steps assume that you are running the perfman executable, but the same steps can easily be modified for the Docker case.

1. Run ./dist/perfman setup, this will deploy the Prometheus Operator and create the service monitors for scraping pipeline and ISB metrics. Pass the -n and/or -j flags to the command if you do not have Numaflow and/or NATS Jetsream ISB running on the cluster
2. In a separate terminal, port-forward the Prometheus service to localhost:9090 with ./dist/perfman portforward -p
3. Deploy the perfman base pipeline: ./dist/perfman pipeline. Allow n minutes to pass in order for metrics to be collected for your desired length of time
4. Collect metrics with: ./dist/perfman -n=<dir-name> -l=<time-period> -m=<metric-group>. dir-name corresponds to the name of the subdirectory that will be created under output/, where each file in this subdirectory will be a CSV file with data pertaining to the metrics associated with the specified metric group. time-period is the amount of time you would like to look back from the current time for the metric collection, and must be less than or equal to n in step 4. metric-group is the metric group to use. The metrics corresponding to a metric group can be found in metrics/metric_groups.go